CORSI be damned!!

The thing that really peaked my interest into advanced analytics in hockey was Chris Boyle's 2013 circular quadrant chart plotting FFclose% vs. playoff success. With the Wild closing in on locking down a playoff spot, and continuing to reside somewhere south of the Mendoza line for shot metrics, many of us band-sledders are feebly trying to hide our apprehension of what that may mean come playoff time.

Chris' post was done 5 years into the analytics era (as I dub as beginning in 2007-08 -€” which is how far back the basic stats are available). It is now 9 years into the era. I decided to re-look at Chris' evaluation to see if it still foretold the same gloom and doom for us Wild fans. (Unfortunately, I am not about to take the time to try to recreate Chris' cool graphic.)

The short, blinders-on answer is: "yes". [sad face] However, as I had downloaded the data from Corsica Hockey (which allows one to download multiple years of data parsed by year in a single download -€” yea!), anyways, I decided to also look at how other metrics do a predicting playoff success. What I found was that, for 9 years of data (144 playoff teams, 135 playoff series), there wasn't an appreciable difference in how well any of the metrics for any single season from Corsica Hockey, save 1, predicted the outcome in the playoffs for a given year. The metrics I looked at where CorsiFor%, CorsiFor%Adj[usted] (for score, venue, and zone starts -€” a simple "close" adjustment is not possible on Corsica), FenwickFor%, FenwickFor%Adj, ScoringChancesFor%, ScoringChancesFor%Adj, GoalsFor%(5on5), GoalsFor%(All situations) and ExpectedGoalsFor%. Note that while Chris Boyle's original data used groupings of <45%, 45 -€” 50%, 50 -€” 55%, and > 55%, the resulting quantities at <45% and >55% were so small (<10% of the values) so as to seem less than ideal to me. Therefore, I reduced the groupings to 3: <50% (which resulted in about 28% of the data), 50 -€” 53% (about 47% of the data), and >53% (about 25% of the data). Unfortunately, I currently don't have the means to get graphics into this editor; and this editor doesn't do tables worth a damn. So, as an example, I am simply going to provide a comparison of CF% and FF%. From there I decided, "what the heck", I might as well look at goals for percentage in all situations (i.e. including PP and PK). Because there is a natural bias with respect to making the playoffs for GF% (that is how teams basically get standing points), I adjusted the grouping values upwards to keep the same breakpoint data percentages. (i.e. the 28% breakpoint was at a GF% value of 51.3%; and the 75% breakpoint was at a GF% value of 54%.) The results (given below) were both surprising and promising:

Made the Playoffs - 144 teams:

<50% CF: 42 teams; 50%-53% CF: 70 teams; >53% CF: 32 teams (5-on-5)

<50% FF: 41 teams; 50%-53% FF: 65 teams; >53% FF: 38 teams (5-on-5)

<51% GF: 40 teams; 51%-54% GF: 68 teams; > 54% GF: 36 teams (all situations)

Made 2nd Round - 72 Teams:

<50% CF: 22 teams; 50%-53% CF: 28 teams; >53% CF: 22 teams (5-on-5)

<50% FF: 21 teams; 50%-53% FF: 29 teams; >53% FF: 22 teams (5-on-5)

<51% GF: 12 teams; 51%-54% GF: 34 teams; > 54% GF: 26 teams (all situations)

Made conference championship - 36 Teams:

<50% CF: 9 teams; 50%-53% CF: 14 teams; >53% CF: 13 teams (5-on-5)

<50% FF: 9 teams; 50%-53% FF: 14 teams; >53% FF: 13 teams (5-on-5)

<51% GF: 5 teams; 51%-54% GF: 15 teams; > 54% GF: 16 teams (all situations)

Made Cup championship - 18 Teams:

<50% CF: 2 teams; 50%-53% CF: 17 teams; >53% CF: 9 teams (5-on-5)

<50% FF: 2 teams; 50%-53% FF: 8 teams; >53% FF: 8 teams (5-on-5)

<51% GF: 1 teams; 51%-54% GF: 17 teams; > 54% GF: 10 teams (all situations)

Stanley Cup Champions - 9 Teams:

<50% CF: 1 team; 50%-53% CF: 2 teams; >53% CF: 6 teams (5-on-5)

<50% FF: 1 team; 50%-53% FF: 3 teams; >53% FF: 5 teams (5-on-5)

<51% GF: 0 teams; 51%-54% GF: 3 teams; > 54% GF: 6 teams (all situations)

A few takeaways for me from this (remembering that all of the 5-on-5 results were very similar to the CF% and FF%):

  • The teams that make the conference finals generally do well in all of the metrics.
  • Adjustments to basic Corsi add little value in predicting playoff success for a single season. (My theory on this is that the uncertainties involved in the factors not accounted for far outweigh the tweaks applied by the adjustments.)
  • Corsica's scoring chances metric does not provide an appreciable improvement over Corsi (for this purpose).
  • There was a nominal, but consistent, improvement in how well all-situation GF% predicted playoff success over what was realized from the 5-on-5 metrics. (Surprise!)

And, the Wild currently sit 2nd in the league in all-situation GF%!!!

But, but, but... what about the Avs from a couple years back ... and the Panthers from last year? (You know, the examples the talking stat-heads love to use as to how unsustainable regular season success doesn't translate to the playoffs.) Well...

For every Avs and Panthers (nearly literally, as I will explain) there is a 07/08 Pens (45.5% CF%, 53.1% GF% - reached Cup final) or 11/12 Rangers (47.8% CF%, 54.95% GF% - reached conference final) or the 08/09 Cup winning Pens! There are 5 such low CF% - high GF% teams that have reached the conference finals or better. There are 6 of the same that have lost out in the first round. One would statistically expect that the first round number (for any given positive metric) would be higher due to the 7 game (max) sample size. Lots of potential for randomness to affect the outcome; and no future games (for the loser) to offset the randomness (as, say, during the regular season). Hence, I'd call that a wash. The Avs and Panthers really aren't a great example.

So CORSI BE DAMNED for 2017! (Of course, I'd much rather be going into the playoffs with a CF% above 50% and a GF% above 55% (the Wild are currently at over 59%). But, this exercise has eased my apprehensions considerably.) Wild for the Cup!!!

The opinions posted here are not those of Hockey Wilderness.